A Low-Carbon, High-Performance Pavement Material Proportioning Intelligent Design System and Method

Abstract

This invention discloses an intelligent design system and method for low-carbon, high-performance paving material proportions, specifically relating to the fields of computer-aided design and intelligent optimization algorithms. It addresses the problem that existing intelligent optimization methods, when dealing with multi-source, strongly coupled constraints, are prone to getting trapped in local optima and lack global optimization capabilities due to design space fragmentation. The method involves analyzing the initial population's violation status of constraints, identifying constraint coupling groups causing spatial fragmentation based on the co-occurrence of violations, dividing and identifying sub-populations corresponding to different fragmentation regions based on these constraint coupling groups, calculating the encoding entropy of each sub-population and its proximity to the average constraint boundary to assess its evolutionary potential, allocating differentiated optimization resources accordingly, and performing independent optimization calculations. Simultaneously, it monitors the convergence trend of each sub-population's evolutionary trajectory. Based on this trend, when it is determined that sub-populations are trending towards similar regions and premature convergence exists, a directional migration of the proportioning scheme across sub-populations is executed, thereby updating the population.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design and intelligent optimization algorithm technology, and more specifically, to an intelligent design system and method for the proportioning of low-carbon, high-performance paving materials. Background Technology

[0002] In the field of road engineering, the mix design of paving materials is a crucial step in ensuring their service performance and engineering economy. This is especially true in areas with complex geological environments and harsh climates, such as karst landforms, where the maintenance materials for transportation infrastructure must meet the dual requirements of low-carbon environmental protection and high performance. Current paving material designs for these objectives must simultaneously satisfy constraints from multiple aspects, including environmental, mechanical, technological, and cost considerations. These constraints include solid waste resource utilization rates, various road performance indicators, construction feasibility windows, and raw material cost limitations. These constraints are intertwined, forming a high-dimensional, nonlinear, and complex design space. To address this complexity, an intelligent approach is adopted, using algorithms to automatically search within the design space for mix design schemes that satisfy all constraints and optimize the overall objectives.

[0003] However, existing intelligent optimization methods that rely on the assumption of continuous space, when faced with the strong coupling effect of the above-mentioned multi-source and rigid constraints, will cause the feasible region that meets all constraints to degenerate from a single connected region into multiple isolated fragmented regions when multiple constraints jointly cut the design space. The algorithm population is easily confined to a certain local feasible fragment for ineffective optimization, and cannot detect and migrate to other isolated regions where better solutions may exist. This leads to design failure or only obtaining local suboptimal solutions, which seriously restricts the reliability and global optimization ability of intelligent design methods in complex engineering scenarios. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a low-carbon, high-performance paving material ratio intelligent design system and method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for intelligent design of low-carbon, high-performance paving material proportions, comprising:

[0007] S1. Obtain the multi-source constraints of the paving materials and construct an initial candidate mix population;

[0008] S2. Analyze the violation states of the initial candidate mix ratio population to the multi-source constraints, and identify and classify the constraint coupling groups that lead to the fragmentation of the design space based on the co-occurrence of the violation states.

[0009] S3. Divide the candidate matching population according to the constraint coupling group, and generate and identify the sub-populations corresponding to different fragmented feasible domains.

[0010] S4. Calculate the coding entropy of the candidate matching schemes within each subpopulation and the closeness to the average constraint boundary, and evaluate the evolutionary potential of each subpopulation based on the coding entropy and the closeness to the average constraint boundary.

[0011] S5. Allocate optimization computing resources to each subpopulation based on the evolutionary potential value and perform independent optimization calculations. At the same time, monitor and analyze the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process.

[0012] S6. Based on the convergence trend of the evolutionary trajectory, perform a directional migration operation for the cross-subpopulation matching scheme and update the candidate matching population.

[0013] Furthermore, S1 includes:

[0014] Obtain and analyze the multi-source constraints to determine the allowable value range of each component of the paving material;

[0015] Multiple candidate ratio schemes are randomly generated within the allowed value range;

[0016] The feasibility of each candidate mix design is determined based on multi-source constraints.

[0017] Candidate pairing schemes with different feasibility judgment results were selected to form an initial candidate pairing population.

[0018] Furthermore, S2 includes:

[0019] For each candidate matching scheme in the initial candidate matching population, record its violation status for each multi-source constraint.

[0020] Statistical analysis of the co-occurrence frequency of any two multi-source constraints being simultaneously violated by the same batch of candidate allocation schemes;

[0021] Multi-source constraints with co-occurrence frequencies exceeding a preset frequency threshold are grouped into a constraint coupling group.

[0022] Furthermore, S3 includes:

[0023] For each constraint coupling group, analyze the overall satisfaction of the multi-source constraint conditions within the constraint coupling group by each candidate ratio scheme in the candidate ratio population.

[0024] Candidate matching schemes that are consistent in overall satisfaction are grouped into the same set;

[0025] Each set generates a subpopulation, and assigns an identifier to the corresponding subpopulation to associate it with its corresponding constraint coupling group.

[0026] Furthermore, S4 includes:

[0027] For each subpopulation, the distribution of coding features of candidate pairing schemes within it is statistically analyzed and the coding entropy is calculated;

[0028] Calculate the distance from each candidate allocation scheme to the boundary of the multi-source constraints, and take the average value to obtain the average constraint boundary proximity.

[0029] The evolutionary potential value of the subpopulation is calculated by combining the encoding entropy with the average constraint boundary proximity according to the preset rules.

[0030] Furthermore, the coding feature distribution of the candidate formulation schemes within the subpopulation is statistically analyzed and the coding entropy is calculated, including: encoding the percentage content of each component in each candidate formulation scheme as a discretized coding bit; statistically analyzing the frequency of occurrence of different discrete values ​​of the coding bit in all coding bits of the subpopulation; and calculating the coding entropy of the subpopulation based on the frequency of occurrence according to the information entropy formula.

[0031] Furthermore, the evolutionary potential value of the subpopulation is calculated by combining the encoding entropy and the average constraint boundary proximity according to preset rules, including: normalizing the encoding entropy and the average constraint boundary proximity respectively; and weighting the normalized encoding entropy and the normalized average constraint boundary proximity according to preset weights. The result is the evolutionary potential value of the subpopulation.

[0032] Furthermore, S5 includes:

[0033] Each subpopulation is assigned an independent number of iterations based on its evolutionary potential value.

[0034] Each subpopulation independently updates and filters candidate matching schemes within the allocated number of iterations.

[0035] During the independent optimization calculation process for each subpopulation, the optimal fitness and coding diversity of each generation of candidate matching schemes are recorded.

[0036] Based on the recorded changes in optimal fitness and coding diversity over multiple generations, the convergence trend of the evolutionary trajectories of each subpopulation was analyzed.

[0037] Furthermore, S6 includes:

[0038] Based on the convergence trend of the evolutionary trajectories of each subpopulation, determine whether there are two subpopulations that tend to be in similar regions in the target space and at least one subpopulation that tends to converge prematurely.

[0039] When the judgment is yes, the candidate matching scheme with the highest fitness is selected from the subpopulation that tends to converge early as the migrating individuals;

[0040] Migrating individuals are added to another subpopulation, and the candidate pairing scheme with the lowest fitness in that other subpopulation is replaced to update the candidate pairing population.

[0041] On the other hand, the present invention provides an intelligent design system for the proportioning of low-carbon, high-performance paving materials, comprising:

[0042] The constraint acquisition module is used to acquire multi-source constraints of paving materials and construct an initial candidate mix population;

[0043] The constraint coupling module is used to analyze the violation state of the initial candidate mix proportion population under multi-source constraints, and to identify and classify constraint coupling groups that lead to design space fragmentation based on the co-occurrence of violation states.

[0044] The subgroup partitioning module is used to partition the candidate matching population according to the constraint coupling group, and generate and identify the subgroups corresponding to different fragmented feasible domains.

[0045] The potential assessment module is used to calculate the coding entropy of the candidate matching schemes within each subpopulation and the closeness to the average constraint boundary, and to assess the evolutionary potential value of each subpopulation based on the coding entropy and the closeness to the average constraint boundary.

[0046] The independent optimization module is used to allocate optimization computing resources to each subpopulation based on the evolutionary potential value and perform independent optimization calculations, while monitoring and analyzing the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process.

[0047] The population update module is used to perform directional migration operations across subpopulations based on the convergence trend of the evolutionary trajectory, and update the candidate matching population.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. By introducing a constraint coupling group identification mechanism based on state co-occurrence violations, it is possible to proactively diagnose the specific causes and structures of design space fragmentation caused by multi-source constraint interactions. This decomposes the originally complex global optimization problem into multiple subspace exploration tasks related to specific constraint combinations. This data-driven problem decomposition strategy enables optimization algorithms to specifically address constraint clusters that cause spatial fragmentation, laying a precise problem cognition foundation for subsequent implementation of partitioned differentiated management. This improves the adaptability and analytical capabilities of intelligent design methods to complex constraint coupling scenarios from the source.

[0050] 2. By dividing the population into subpopulations corresponding to different fragmented regions and dynamically assessing their evolutionary potential by comprehensively encoding diversity and boundary proximity, differentiated resource allocation and independent parallel optimization are implemented. This achieves efficient tilting of computational resources toward more promising search directions. Crucially, by monitoring the convergence trend of each subpopulation and triggering directional migration based on spatial proximity at appropriate times, information exchange between fragments of different feasible domains is effectively simulated and promoted. This proactively breaks the search stagnation caused by spatial isolation, guides the optimization process to cross local fragment barriers, and thus systematically enhances the robustness of global exploration in fragmented feasible domains and the overall quality of the final solution. Attached Figure Description

[0051] Figure 1 This is a flowchart of an intelligent design method for the proportion of low-carbon, high-performance paving materials according to the present invention.

[0052] Figure 2 This is a schematic diagram of the structure of an intelligent design system for the proportioning of low-carbon, high-performance paving materials according to the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example 1: Figure 1 This invention presents an intelligent design method for the proportioning of low-carbon, high-performance paving materials, comprising:

[0055] S1. Obtain the multi-source constraints of the paving materials and construct an initial candidate mix population;

[0056] S2. Analyze the violation states of the initial candidate mix ratio population to the multi-source constraints, and identify and classify the constraint coupling groups that lead to the fragmentation of the design space based on the co-occurrence of the violation states.

[0057] S3. Divide the candidate matching population according to the constraint coupling group, and generate and identify the sub-populations corresponding to different fragmented feasible domains.

[0058] S4. Calculate the coding entropy of the candidate matching schemes within each subpopulation and the closeness to the average constraint boundary, and evaluate the evolutionary potential of each subpopulation based on the coding entropy and the closeness to the average constraint boundary.

[0059] S5. Allocate optimization computing resources to each subpopulation based on the evolutionary potential value and perform independent optimization calculations. At the same time, monitor and analyze the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process.

[0060] S6. Based on the convergence trend of the evolutionary trajectory, perform a directional migration operation for the cross-subpopulation matching scheme and update the candidate matching population.

[0061] S1. Obtain the multi-source constraints of the paving materials and construct an initial candidate mix population, specifically implemented as follows:

[0062] Obtaining multi-source constraints for paving materials includes environmental constraints, mechanical performance constraints, construction process constraints, and raw material cost constraints. Environmental constraints refer to mandatory requirements on the content of specific components in paving materials. For example, the content of industrial solid waste such as phosphogypsum or fly ash must not be less than 30%, a percentage derived from explicit provisions in the environmental protection regulations of the project location. Mechanical performance constraints refer to the minimum or maximum limits for various road performance indicators of the paving material after molding, such as a Marshall stability of not less than 8000 Newtons and a flow value between 2 mm and 4 mm. These specific limits are derived from the technical specifications in the design drawings or industry-recognized technical standards. Construction process constraints refer to the range of process parameters that ensure the smooth mixing, paving, and compaction of materials. For example, the mixing temperature of asphalt mixtures must be between 150°C and 170°C, and the compaction degree must not be less than 96%. These ranges are derived from the typical operating parameters of the construction equipment used and summaries of engineering practice experience. Raw material cost constraints refer to the requirement that the total cost of a single unit of material must be controlled below a specific amount, which is derived from the project's budget documents. All constraints are manually read from the corresponding documents and reports and input into the computer system as key-value pairs, forming a structured list of constraints.

[0063] The permissible value ranges for each component of the pavement material are determined based on multi-source constraints. The components of the pavement material include asphalt, coarse aggregate, fine aggregate, mineral fillers, and industrial solid waste substitutes. The method for determining the permissible value range involves mathematically transforming each specific constraint into a linear or nonlinear inequality constraint on the content of one or more components. For example, the solid waste resource utilization rate constraint directly provides a lower limit for the phosphogypsum content, such as 30%. The asphalt content constraint provides a range based on gradation type and engineering experience, such as 4.5% to 5.5%. The aggregate gradation constraint is transformed into the upper and lower limits of the sieve passing rate. By comprehensively considering all the inequality constraints imposed on each component by all constraints, the intersection of the solution sets of these inequalities represents the permissible value range for that component while simultaneously satisfying all multi-source constraints. This process is completed automatically by a computer program. The program solves the inequalities obtained from the analysis and uses linear programming or interval solving algorithms to calculate the independent allowable range of values ​​for each component. This range is represented by a minimum percentage value and a maximum percentage value.

[0064] Multiple candidate formulation schemes are randomly generated within the allowed value range. Random generation employs uniform random sampling, where a computer program independently selects a specific percentage value for each component within its allowed range. All randomly selected values ​​for each component together constitute a complete candidate formulation scheme. The sum of the percentage content of each component in each candidate formulation scheme is 100%. To ensure that the initial sampling broadly covers the entire possible solution space, the number of randomly generated candidate formulation schemes is set to a large number, for example, 10,000 random schemes. Each scheme is stored in the form of a vector, where the value of each dimension represents the specific percentage content of a particular component.

[0065] The feasibility of each candidate mix design is determined based on multi-source constraints. The process involves substituting the specific component data of each candidate mix design into each specific constraint inequality within the multi-source constraints for verification. For example, for a given design, the feasibility is calculated to determine if the actual phosphogypsum content is greater than or equal to the lower limit of environmental constraints, if the estimated Marshall stability is greater than or equal to the minimum performance constraint, and if the estimated mixing temperature is within the temperature range required by the construction process constraints. Each constraint generates a Boolean judgment result indicating whether it has been violated, i.e., a violation status. If the design fully satisfies a constraint, the violation status is recorded as "not violated"; if the design does not satisfy a constraint, the violation status is recorded as "violated." A design needs to undergo a complete evaluation of all multi-source constraints, ultimately yielding a violation status vector. This vector records the satisfaction status of the design relative to each specific constraint.

[0066] Candidate mix design schemes with different feasibility assessment results are selected to form an initial candidate mix design population. The selection is based on the overall feasibility assessment results of the candidate mix design schemes, i.e., categorized according to their violation state vectors. The classification criteria are based on the number and type of violated constraints. For example, schemes that fully satisfy all constraints are classified into category one, schemes that violate only one construction technology constraint into category two, schemes that violate one mechanical performance constraint into category three, and schemes that violate both environmental and cost constraints into category four. A certain number of schemes are randomly selected from each category according to a preset ratio, for example, 100 schemes from category one, 200 schemes from category two, 150 schemes from category three, and 50 schemes from category four, ensuring that the final total number of selected schemes is 500, and that these schemes cover different feasibility types. These 500 candidate mix design schemes with different feasibility assessment results constitute the initial candidate mix design population for subsequent steps. The initial population constructed in this way not only contains fully feasible potential solutions, but also some infeasible solutions that are close to the boundary of the feasible domain, providing a diverse sample for subsequent analysis and design of spatially fragmented structures.

[0067] S2. Analyze the violation states of the initial candidate mix proportion population to the multi-source constraints, and identify and classify the constraint coupling groups that lead to the fragmentation of the design space based on the co-occurrence of the violation states. The specific implementation is as follows:

[0068] For each candidate formulation in the initial candidate formulation population, its violation status against each multi-source constraint is recorded. This process reads the specific component data of each candidate formulation in the initial candidate formulation population and the specific inequalities or range requirements of each constraint in the multi-source constraints. For a specific candidate formulation in the initial candidate formulation population, its component data is sequentially substituted into each multi-source constraint for verification calculation. The verification calculation produces a binary logic result: violation or no violation. For example, for an environmental constraint requiring a phosphogypsum content of no less than 30%, the actual percentage content of phosphogypsum in the candidate formulation is calculated. If the content is greater than or equal to 30%, the violation status of this multi-source constraint is recorded as no violation; if the content is less than 30%, the violation status is recorded as violation. For a mechanical performance constraint requiring a Marshall stability of at least 8000 Newtons, a pre-defined empirical regression model is used to estimate the predicted Marshall stability value of the candidate formulation based on its component data. If the predicted value is greater than or equal to 8000 Newtons, the violation state is recorded as "not violated"; otherwise, it is recorded as "violated." This process is repeated for all multi-source constraints, generating a violation state vector for the current candidate formulation. The length of this vector equals the total number of multi-source constraints, and each value in the vector corresponds to a violation state of a specific multi-source constraint. This verification and recording process is repeated for all candidate formulations in the initial candidate formulation population, ultimately resulting in a violation state matrix. Rows in this matrix correspond to each candidate formulation, and columns correspond to each multi-source constraint. Each element in the matrix is ​​a binary logical value, recording whether a specific candidate formulation satisfies a specific multi-source constraint.

[0069] After obtaining the complete violation state matrix, the co-occurrence frequency of any two multi-source constraints being simultaneously violated by the same batch of candidate pairing schemes is calculated. The statistical process first requires selecting any two different multi-source constraints from all the constraints to form a constraint pair. For a selected constraint pair, all rows of the violation state matrix (i.e., all candidate pairing schemes) are traversed, and the violation state of each candidate pairing scheme for both multi-source constraints is checked. The number of candidate pairing schemes that simultaneously violate both multi-source constraints is counted; that is, the number of schemes where both multi-source constraints are violated in the same candidate pairing scheme. This number is the absolute co-occurrence frequency of the two multi-source constraints. The co-occurrence frequency is calculated by dividing the absolute co-occurrence frequency by the total number of candidate pairing schemes in the initial candidate pairing population whose violation state is not entirely unviolated, resulting in a frequency value between 0 and 1. This frequency value reflects the correlation strength of the simultaneous failure of two multi-source constraints in the sampling space represented by the current initial candidate pairing population. The above statistics and calculations need to be performed on all possible combinations of multi-source constraint pairs to calculate a unique co-occurrence frequency value for each pair. For example, if there are 10 multi-source constraints, then the co-occurrence frequencies of 45 different constraint pairs need to be calculated.

[0070] Multi-source constraints with a co-occurrence frequency exceeding a preset frequency threshold are grouped into a constraint coupling group. The preset frequency threshold is a threshold value used to determine whether co-occurrence is significant, and its setting is based on the statistical distribution characteristics of the co-occurrence frequency. One method is to calculate the average and standard deviation of the co-occurrence frequencies of all constraint pairs, and set the preset frequency threshold as the average plus one standard deviation. Another method is to sort the co-occurrence frequency values ​​of all constraint pairs from largest to smallest, and use the co-occurrence frequency value at the 80th percentile of the sorted values ​​as the preset frequency threshold. After determining the preset frequency threshold, all constraint pairs are traversed, and constraint pairs with a co-occurrence frequency value greater than the preset frequency threshold are selected. Based on these selected high co-occurrence frequency constraint pairs, a community detection algorithm from graph theory is used to divide the constraint coupling groups. Specifically, each multi-source constraint is considered a node in the graph. If the co-occurrence frequency of two multi-source constraints is greater than the preset frequency threshold, an edge is added between the two nodes, and the weight of the edge is the co-occurrence frequency value. Subsequently, a community detection algorithm, such as the Louvain algorithm based on modularity optimization, is run on this weighted graph to divide it into several node communities with tightly connected internal connections and sparse external connections. Each divided node community contains multi-source constraints, which are identified as a constraint coupling group. The multi-source constraints within each constraint coupling group tend to be violated simultaneously by the same batch of candidate allocation schemes, indicating that they work together in the design to define the boundary of a specific, fragmented feasible region isolated from other regions in the design space. After partitioning, each constraint coupling group is assigned a unique group identifier, and all the specific multi-source constraints contained within the group are recorded for use in subsequent steps.

[0071] S3. Based on the constraint coupling group, divide the candidate matching population into subpopulations, generate and identify subpopulations corresponding to different fragmented feasible regions. The specific implementation is as follows:

[0072] For each constraint coupling group, the overall satisfaction of the multi-source constraints within that group is analyzed for each candidate pairing scheme in the candidate pairing population. The analysis process uses the list of constraint coupling groups and the violation state matrix of the candidate pairing population as input. For a specific constraint coupling group, all multi-source constraints contained in that group are extracted from the constraint coupling group list. The violation state of each candidate pairing scheme against these extracted multi-source constraints is extracted from the violation state matrix, forming a sub-matrix. The goal of the analysis is to calculate an overall satisfaction evaluation value for each candidate pairing scheme regarding that constraint coupling group. One calculation method is to count the number of violations of all multi-source constraints within the constraint coupling group by the candidate pairing scheme. For example, if a constraint coupling group contains 5 multi-source constraints, for a specific candidate pairing scheme, the corresponding row in the sub-matrix is ​​examined, and the number of violations is counted; this number is the overall satisfaction evaluation value, with a value of 0 indicating complete satisfaction and a value of 5 indicating complete violation. Another calculation method is a weighted violation score, assigning a weight to each multi-source constraint within the constraint coupling group. The weights are set based on the statistical contribution of the multi-source constraint to infeasibility in the initial candidate matching population. Specifically, for a specific multi-source constraint within the constraint coupling group, the total number of violations (i.e., violations) in a submatrix of the violation state matrix is ​​counted, denoted as the individual violation frequency. Then, the sum of the co-occurrence frequencies of this multi-source constraint and other multi-source constraints within the constraint coupling group is calculated. The individual violation frequency of this multi-source constraint is multiplied by the sum of the co-occurrence frequencies to obtain the initial influence factor of the constraint. After calculating the initial influence factors for all multi-source constraints within the constraint coupling group, all initial influence factors are normalized so that their sum is 1. The normalized result is the final weight value for each multi-source constraint. When calculating the weighted score, the violation state of each multi-source constraint for the candidate matching scheme is multiplied by the corresponding final weight value of the multi-source constraint (violation counted as 1, no violation counted as 0), and then summed to obtain the weighted score as the overall satisfaction evaluation value. This analysis process performs the above calculations for each candidate matching scheme in the candidate matching population, for the currently processed constraint coupling group. Ultimately, it generates a list for the current constraint coupling group, where each item records a candidate matching scheme and its corresponding overall satisfaction evaluation value. This process is repeated for each identified constraint coupling group, thereby establishing a mapping between the overall satisfaction relationship between candidate matching schemes and each constraint coupling group.

[0073] After obtaining the overall satisfaction evaluation values ​​of all candidate allocation schemes relative to each constraint coupling group, candidate allocation schemes with consistent overall satisfaction are grouped into the same set. The classification operation is based on the overall satisfaction evaluation value. For a specific constraint coupling group, all candidate allocation schemes in the overall satisfaction evaluation value list corresponding to that constraint coupling group are grouped according to the magnitude of the overall satisfaction evaluation value. One or more classification thresholds are set, and candidate allocation schemes are divided into different category sets based on the comparison relationship between the overall satisfaction evaluation value and the classification threshold. For example, for a constraint coupling group that uses the number of violations as the overall satisfaction evaluation value, two classification thresholds can be set. The first classification threshold is the boundary value used to distinguish between complete satisfaction and partial violation, and the second classification threshold is the boundary value used to distinguish between partial violation and severe violation. For example, the first classification threshold can be set to 0, and the second classification threshold can be set to 2. Candidate pairing schemes with an overall satisfaction rating of 0 are categorized into set A, indicating complete fulfillment of the boundary definition for the constraint coupling group; candidate pairing schemes with an overall satisfaction rating greater than 0 and less than or equal to 2 are categorized into set B, indicating partial violation; and candidate pairing schemes with an overall satisfaction rating greater than 2 are categorized into set C, indicating severe violation. The classification thresholds can be set based on the statistical distribution of the overall satisfaction ratings. One method is to calculate the mean and standard deviation of the overall satisfaction ratings of all candidate pairing schemes for the current constraint coupling group, setting the first classification threshold as the mean minus one standard deviation, and the second classification threshold as the mean plus one standard deviation. Another method is to set the thresholds based on quantiles. For example, all candidate pairing schemes are sorted in ascending order of their overall satisfaction ratings, and the overall satisfaction rating of the candidate pairing scheme at the 33% position is taken as the first classification threshold, and the overall satisfaction rating of the candidate pairing scheme at the 66% position is taken as the second classification threshold. For constrained coupling groups that use weighted scoring as the overall satisfaction evaluation value, the principle for setting the classification threshold is the same, determined based on the distribution characteristics of their score values. After the classification operation is completed, for a constrained coupling group, several mutually exclusive sets of candidate allocation schemes will be obtained, and all candidate allocation schemes in each set have a consistent overall satisfaction category. This classification process is performed independently for each constrained coupling group; therefore, a candidate allocation scheme may belong to different categories in different constrained coupling groups simultaneously.

[0074] After classifying the candidate allocation schemes for all constraint coupling groups, a subpopulation is generated for each set, and an identifier is assigned to each subpopulation to associate it with its corresponding constraint coupling group. Generating a subpopulation involves extracting all candidate allocation schemes belonging to the same set, copying their complete component data, and forming an independent, smaller group of candidate allocation schemes; this group is a subpopulation. The generation of subpopulations is set-based, and sets are divided according to the overall satisfaction of a specific constraint coupling group. Therefore, each subpopulation is naturally associated with a specific constraint coupling group and a specific overall satisfaction category. The purpose of assigning identifiers to subpopulations is to trace their origin and characteristics in subsequent steps. The identifier encoding rule is designed to include three parts: the constraint coupling group number, the overall satisfaction category code, and the subpopulation sequence number. For example, the constraint coupling group number can be represented by the letter G followed by a number, such as G01 and G02. The overall satisfaction category code can be represented by the letter C followed by a number, such as C0 representing complete satisfaction, C1 representing partial violation, and C2 representing severe violation. Subpopulation sequence numbers are used to distinguish multiple subsets within the same category under the same constraint coupling group, which may be generated due to different classification thresholds. They can be represented by the letter S followed by a number, such as S01. Therefore, a complete subpopulation identifier could be G01_C0_S01, indicating that the subpopulation originates from the first subset of candidate proportioning schemes that fully satisfy constraint coupling group G01. In the program implementation, this identifier is stored as a string attribute bound to the subpopulation object. All generated subpopulations are collected into a global subpopulation list. Each subpopulation object contains data on its member candidate proportioning schemes, its size, and its unique identifier. Through this partitioning and identification mechanism, the original candidate proportioning population is systematically decomposed into multiple subpopulations. Each subpopulation corresponds to a potential design space fragmentation region dominated by a specific constraint coupling group, and the candidate proportioning schemes within a subpopulation have similar states on the boundary defined by that constraint coupling group. This lays the structural foundation for subsequent differentiated evolution and monitoring.

[0075] S4. Calculate the coding entropy of the candidate matching schemes within each subpopulation and its closeness to the average constraint boundary, and evaluate the evolutionary potential of each subpopulation based on the closeness of the coding entropy to the average constraint boundary. Specifically, this is implemented as follows:

[0076] For each subpopulation, the distribution of coding features of candidate formulations within it is statistically analyzed, and the coding entropy is calculated. Specifically, this involves reading the component data of all candidate formulations in the subpopulation; each candidate formulation consists of the percentage content of multiple components. First, a coding operation is performed, discretizing the continuous percentage values ​​of each component into a pre-defined finite number of levels. For example, for a specific component such as asphalt content, its allowed range is 4.5% to 5.5%. This range is evenly divided into ten discrete intervals, such as 4.5% to 4.6% for the first interval, 4.6% to 4.7% for the second interval, and so on. Each discrete interval corresponds to a discretized coding value, for example, represented by numbers 1 to 10. For a candidate formulation in a subpopulation, if its asphalt content is 4.63%, it falls into the second interval, and the coding value for this component in this formulation is 2. This discretization coding operation is performed on each component in the formulation. Finally, each candidate formulation is represented as a sequence of multiple discrete coding values, with each coding value at a position called a coding bit. After encoding all schemes, for each subpopulation, the frequency of different discrete encoded values ​​at all encoding positions is counted. For each encoding position, the frequency of each discrete encoded value at that position is counted for all candidate pairing schemes within the subpopulation, and the proportion of each discrete encoded value's frequency to the total number of schemes in the subpopulation is calculated. This proportion is the frequency of that discrete encoded value at that encoding position. After obtaining the frequencies of all possible discrete encoded values ​​at all encoding positions, the encoding entropy of the subpopulation is calculated using the information entropy formula. The calculation process of the information entropy formula is as follows: for each encoding position, the frequency of each discrete encoded value at that position is multiplied by the logarithm of its respective frequency to base 2, and then all these products are summed and the negative value is taken to obtain the entropy value of that encoding position. Finally, the entropy values ​​of all encoding positions are arithmetically averaged, and the average value is the encoding entropy of the subpopulation. The larger the encoding entropy value, the more uniform the distribution of the encoding features of the candidate pairing schemes within the subpopulation, and the higher the diversity.

[0077] The distance from each candidate mix design to the boundary of the multi-source constraints is calculated, and the average distance is taken to obtain the average constraint boundary proximity. The calculation process uses candidate mix designs and multi-source constraints in the subpopulation as input. For a candidate mix design in the subpopulation, its distance to the boundary of each relevant multi-source constraint needs to be calculated. Multi-source constraints are usually expressed as inequality restrictions on one or a set of variables. For example, a constraint may require that the porosity of the asphalt mixture is no greater than 5%. For such inequality constraints, the boundary is the plane with a porosity of 5%. To calculate the distance of a candidate mix design to this boundary, the predicted porosity value needs to be calculated based on the component data of the design using a pre-set empirical model or theoretical formula. Then, calculate the absolute difference between the predicted porosity value and the boundary value of 5%, and divide this difference by a normalization factor. This normalization factor can be set to the typical range of values ​​for this type of performance index. For example, if the typical range of porosity variation is 2% to 8%, then the normalization factor can be set to 6%, thus obtaining a dimensionless relative distance. For another type of constraint, such as requiring the powder-to-binder ratio to be within a range, such as 0.8 to 1.2, its boundary is defined by both the upper and lower limits. In this case, it is necessary to calculate the distance between the predicted powder-to-binder ratio of the candidate formulation and the nearest boundary. If the predicted value is within the range, the distance is 0; if the predicted value is less than the lower limit of 0.8, the distance is the lower limit minus the predicted value; if the predicted value is greater than the upper limit of 1.2, the distance is the predicted value minus the upper limit. Similarly, this distance value needs to be normalized by dividing by the typical range of values ​​for this constraint, 1.2 minus 0.8, i.e., 0.4. Repeat the above distance calculation process for all multi-source constraints involved in the candidate formulation to obtain a set of normalized distance values. The arithmetic mean of these distance values ​​is the comprehensive boundary distance of the candidate matching scheme. Finally, the arithmetic mean of the comprehensive boundary distances of all candidate matching schemes within the subpopulation is calculated; this mean is the average constraint boundary proximity of the subpopulation. The smaller the average constraint boundary proximity, the closer the candidate matching schemes within the subpopulation are to the feasible region boundary defined by the multi-source constraints.

[0078] The evolutionary potential of a subpopulation is calculated by combining the encoded entropy with the average constraint boundary proximity according to preset rules. The preset rules include two core steps: normalization and weighted summation. First, the encoded entropy and the average constraint boundary proximity are normalized separately. Normalization needs to be performed based on the global distribution of the encoded entropy and average constraint boundary proximity of all subpopulations to be evaluated. Specifically, after calculating the encoded entropy of all subpopulations, the maximum and minimum values ​​of all encoded entropies are identified. For the encoded entropy of a specific subpopulation, its normalized value is calculated by subtracting the minimum value of all encoded entropies from the encoded entropy, and then dividing the difference by the difference between the maximum and minimum values ​​of all encoded entropies. Similarly, after calculating the average constraint boundary proximity of all subpopulations, the maximum and minimum values ​​of all average constraint boundary proximity are identified. For the average constraint boundary proximity of a specific subpopulation, its normalized value is calculated by subtracting the minimum value among all average constraint boundary proximity values ​​from the average constraint boundary proximity value, and then dividing the difference by the difference between the maximum and minimum values ​​among all average constraint boundary proximity values. Through this process, the encoding entropy and average constraint boundary proximity of each subpopulation are mapped to a numerical range of 0 to 1. After normalization, a weighted summation is performed. The weighted summation requires pre-setting two weight coefficients: a weight coefficient for encoding entropy and a weight coefficient for average constraint boundary proximity. The rule for setting the weight coefficients is based on the relative importance of the two indicators to the evolutionary potential. A common rule is to first analyze the main objective of the current optimization task. If the early exploration phase focuses on expanding the search range and discovering new regions, then assign a higher weight coefficient to encoding entropy and a lower weight coefficient to average constraint boundary proximity. If the later convergence phase focuses on fine-tuning in advantageous regions, then assign a higher weight coefficient to average constraint boundary proximity and a lower weight coefficient to encoding entropy. The weighting coefficients range from 0 to 1, and the sum of any two weighting coefficients is 1. In practice, multiple weighting coefficient schemes can be set. For example, in the exploration phase, the scheme might be a coding entropy weight of 0.7 and an average constraint boundary proximity weight of 0.3; in the development phase, the scheme might be a coding entropy weight of 0.4 and an average constraint boundary proximity weight of 0.6. The appropriate weighting scheme is dynamically selected based on the optimization progress. After setting the weighting coefficients, the evolutionary potential value of a subpopulation is calculated by multiplying the subpopulation's normalized coding entropy by its weighting coefficient, and then adding the subpopulation's normalized average constraint boundary proximity multiplied by its weighting coefficient. The result is a value between 0 and 1, representing the subpopulation's evolutionary potential value. A higher evolutionary potential value indicates that the subpopulation is considered to have a higher expected return in subsequent optimization calculations, and therefore should be allocated more computational resources. After calculating the evolutionary potential values ​​of all subpopulations, a list is created for use in subsequent steps.

[0079] S5. Allocate optimization computing resources to each subpopulation based on its evolutionary potential value and perform independent optimization calculations. Simultaneously, monitor and analyze the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process. The specific implementation is as follows:

[0080] Based on the proportion of evolutionary potential values ​​of each subpopulation, an independent number of iterations is allocated to each subpopulation. The allocation process takes a list of all subpopulations and their corresponding evolutionary potential values ​​as input. First, the sum of the evolutionary potential values ​​of all subpopulations is calculated. Then, for a specific subpopulation, the number of independent iterations allocated to it is calculated by dividing the evolutionary potential value of that subpopulation by the sum of the evolutionary potential values ​​of all subpopulations, thus obtaining the proportion of computing resources that subpopulation should occupy. Multiplying this proportion of computing resources by a preset global limit on the total number of iterations yields the number of independent iterations allocated to that subpopulation. For example, if there are three subpopulations with evolutionary potential values ​​of 0.5, 0.3, and 0.2, the sum of their evolutionary potential values ​​is 1.0. The preset global limit on the total number of iterations is 1000. The first subpopulation is allocated (0.5 / 1.0) × 1000 = 500 independent iterations; the second subpopulation is allocated (0.3 / 1.0) × 1000 = 300; and the third subpopulation is allocated (0.2 / 1.0) × 1000 = 200. The preset upper limit for the total number of global iterations is based on available computation time and desired optimization accuracy. This limit is set through algorithm pre-running. Specifically, the optimization algorithm is run on a representative standard test problem, and the change in solution quality with increasing iteration count is observed. The average number of iterations required to achieve a satisfactory solution quality is recorded, and this average number of iterations is multiplied by a safety factor, such as 1.5. The resulting product is the preset upper limit for the total number of global iterations. After each subpopulation receives its allocated number of independent iterations, this number serves as a termination condition for its independent optimization process.

[0081] Each subpopulation independently updates and filters candidate resource allocation schemes within its allocated number of independent iterations. The independent optimization process initiates a separate computational task for each subpopulation. The update and filtering operations employ a genetic algorithm. Specifically, all candidate resource allocation schemes currently included in the subpopulation are used as the initial parent population. In each iteration, the fitness of each candidate resource allocation scheme in the parent population is calculated first. Fitness is calculated based on multi-source constraints and performance objectives. The fitness function is constructed by weighted summing of the performance values ​​of each candidate resource allocation scheme, such as road performance predictions, carbon emission estimates, and cost estimates, into a comprehensive score. Simultaneously, a numerical penalty is applied to violations of multi-source constraints. This penalty is calculated by multiplying each violated constraint by a penalty coefficient based on its severity, summing all penalty values, and subtracting the sum from the comprehensive score. The final value is the fitness; a higher fitness indicates a better scheme. Then, a roulette wheel selection process is used to choose candidate pairing schemes from the parent population based on their fitness, with higher-fitting schemes having a higher probability of being selected for reproduction. Next, a single-point crossover operation is performed on the selected schemes. This operation randomly selects two parent schemes, randomly selects a coding position, and swaps the coding values ​​of all components after that position, generating two new offspring schemes. Subsequently, a basic bit mutation operation is performed on the offspring schemes. This operation randomly changes the coding value of a component in the offspring scheme to another allowed discrete coding value with a small probability. Through selection, crossover, and mutation, a new generation of candidate pairing schemes is generated. Finally, from the parent and offspring schemes, a number of superior schemes, equal to the size of the original subpopulation, are selected based on their fitness ranking to serve as the parent population for the next iteration. This process is repeated until the number of independent iterations allocated to that subpopulation is reached. The independent optimization calculation processes of each subpopulation are isolated from each other and do not interfere with each other.

[0082] During the independent optimization process of each subpopulation, the optimal fitness and coding diversity of each generation of candidate pairing schemes are recorded. This recording operation is performed after the selection step of each iteration. For optimal fitness, its value is the maximum fitness among all schemes selected in the current iteration's candidate pairing scheme group. This maximum value is recorded in a historical optimal fitness sequence corresponding to that subpopulation. Coding diversity is calculated by statistically analyzing the distribution of different discrete coding values ​​at all coding positions for the current generation of candidate pairing schemes. A specific quantitative indicator of coding diversity is the average Hamming distance between all pairs of schemes in the group. The Hamming distance is calculated by comparing the coding sequences of two candidate pairing schemes and counting the number of positions where their coding values ​​differ at the same coding position. The Hamming distance between all possible pairs of schemes in the current generation is calculated, and the arithmetic mean of all Hamming distances is taken. This arithmetic mean is the coding diversity value for the current generation. This coding diversity value is recorded in a historical coding diversity sequence corresponding to that subpopulation. Therefore, as the independent optimization computation proceeds, each subpopulation will accumulate two sequences of length equal to the number of iterations performed: the historical best fitness sequence and the historical coding diversity sequence.

[0083] Based on the recorded changes in optimal fitness and coding diversity across multiple generations, the convergence trend of the evolutionary trajectory of each subpopulation is analyzed. The analysis process is performed uniformly after the subpopulation's independent optimization calculations have reached the allocated number of independent iterations. Generating the convergence trend of the evolutionary trajectory requires defining specific convergence criteria. One approach is to define two types: premature convergence and asymptotic convergence. For judging premature convergence, two thresholds need to be set: a fitness stagnation threshold and a coding diversity lower bound threshold. The fitness stagnation threshold is a preset positive integer used to determine whether the historical optimal fitness sequence has stagnated. The coding diversity lower bound threshold is a preset small value used to determine whether the coding diversity is too low. During the analysis, the fitness stagnation generation number is first calculated; this refers to the number of iterations in the historical optimal fitness sequence where the maximum value has remained unchanged. Then, the current coding diversity is obtained, which is the last value of the historical coding diversity sequence. If the fitness stagnation generation number exceeds the preset fitness stagnation threshold, and the current coding diversity is lower than the preset coding diversity lower bound threshold, then the evolutionary trajectory convergence trend of the subpopulation is determined to be premature convergence. To determine whether a population is in a gradual convergence state, a convergence rate threshold needs to be set. This threshold is a pre-defined, very small positive number used to determine whether the rate of increase in optimal fitness approaches zero. During analysis, the rate of change of the historical optimal fitness sequence over the most recent generations is calculated. If the absolute value of this rate of change is less than the pre-defined convergence rate threshold, and the historical coding diversity sequence shows a stable or slowly decreasing trend, then the evolutionary trajectory of the subpopulation is determined to be in a gradual convergence state. If neither the conditions for premature convergence nor gradual convergence are met, then the evolutionary trajectory of the subpopulation is determined to be in an active exploration state. The specific values ​​of the fitness stagnation threshold, coding diversity lower bound threshold, and convergence rate threshold are calibrated through algorithm pre-running. The calibration method involves running the algorithm multiple times on a typical test problem, observing the convergence behavior of the algorithm under different parameter combinations, and recording the parameter values ​​that accurately distinguish between premature convergence, gradual convergence, and active exploration. These parameter values ​​are then used as the final thresholds. The convergence trend of the evolutionary trajectory generated for each subpopulation, along with its current best candidate matching scheme and population state, will be used as a complete output to guide subsequent migration decisions.

[0084] S6. Based on the convergence trend of the evolutionary trajectory, perform a directional migration operation for the cross-subpopulation matching scheme to update the candidate matching population. The specific implementation is as follows:

[0085] Based on the convergence trend of the evolutionary trajectories of each subpopulation, it is determined whether there are two subpopulations tending towards similar regions in the target space, and at least one subpopulation tending towards premature convergence. The determination process uses the current state of all subpopulations as input, which includes the convergence trend of the evolutionary trajectory of each subpopulation, its historical best fitness sequence, its current generation of candidate formulation schemes, and the coordinates of each candidate formulation scheme in the target space. The target space is a multi-dimensional space, where each dimension corresponds to a performance index or optimization objective of the paving material, such as Marshall stability, dynamic modulus, fatigue life, carbon emissions, and material cost. The coordinates of a candidate formulation scheme in the target space are obtained by applying a predetermined performance prediction model and carbon emission estimation model to the component data of that scheme; all coordinates are dimensionless normalized values. The first step of the determination process is to screen out all subpopulations whose evolutionary trajectory convergence trends are marked as premature convergence trends, forming a set of premature subpopulations. The second step in the judgment process is to calculate the current position of each subpopulation in the target space for each precocious subpopulation set. The current position is determined by using the coordinates of the candidate matching scheme with the highest fitness in the current generation of that subpopulation as a representative point in the target space. The third step is to calculate the distance between the representative point of the precocious subpopulation and the representative points of all other non-precocious convergent subpopulations. The distance is calculated using the Euclidean distance formula. Specifically, for two representative points, first, the difference in their coordinates in each dimension of the target space is calculated, each difference is squared, then the squared differences in all dimensions are summed, and finally, the square root of the sum is taken. The resulting value is the Euclidean distance between the representative points of the two subpopulations. The fourth step is to set a distance threshold to determine whether two subpopulations tend to converge to similar regions in the target space. The distance threshold is calibrated through algorithm pre-running. The calibration method involves running the complete algorithm flow on a typical test problem, recording the Euclidean distances between all pairwise representative points of all subpopulations, calculating the mean and standard deviation of all Euclidean distances, and setting the distance threshold as the mean minus one standard deviation. If the Euclidean distance between a representative point of a precocious subpopulation and a representative point of a non-precocious subpopulation is less than or equal to the preset distance threshold, then the two subpopulations are determined to tend towards a similar region in the target space. The output of the entire judgment process is "yes" when at least one pair of subpopulations satisfying the above Euclidean distance condition is found, and the evolutionary trajectory convergence trend of at least one of the subpopulations is a precocious convergence trend.

[0086] When the determination is yes, the candidate matching scheme with the highest fitness is selected as the migrant individual from the subpopulation that tends towards premature convergence. The selection operation targets the subpopulation that has been determined to tend towards premature convergence. First, the candidate matching scheme population for the current generation of this subpopulation is obtained. Then, the fitness value of each candidate matching scheme in the population is read. The fitness value has been calculated and stored in the previous independent optimization calculation process. Next, the maximum value among all fitness values ​​is found. Finally, the candidate matching scheme with the maximum fitness value is selected as the migrant individual. If multiple candidate matching schemes have the same highest fitness value, one is randomly selected as the migrant individual using a random number generator. The selected migrant individual retains its complete component data and its coordinate information in the target space.

[0087] The process involves adding migrating individuals to another subpopulation and replacing the candidate pairing scheme with the lowest fitness in that subpopulation to update the candidate pairing scheme population. This other subpopulation refers to a non-precocious subpopulation that successfully paired with a precocious convergent subpopulation during the decision-making process and tends towards a similar region in the target space. The first step of the update operation is to add the migrating individual as a new scheme to the current candidate pairing scheme population of the other subpopulation. At this point, the population size is temporarily increased by one. The second step of the update operation is to identify the candidate pairing scheme with the lowest fitness value from this temporarily expanded population. The lowest fitness value is determined by comparing the fitness values ​​of all schemes in the population and finding the minimum value. If multiple schemes have the same lowest fitness value, one is randomly selected using a random number generator. The third step of the update operation is to permanently remove this candidate pairing scheme with the lowest fitness value from the population. After adding migrating individuals and removing the individual with the lowest fitness, the candidate pairing scheme population of the other subpopulation is restored to its original size. This series of operations constitutes a complete migration event, which injects a high-quality solution from a prematurely convergent region into a population that is searching in a similar region but has not yet stagnated, while eliminating the worst solution in the target population. This migration event is recorded, including the source subpopulation identifier of the migrating individual, the target subpopulation identifier, and the generation in which the migration occurred. After the migration update, the two related subpopulations will continue subsequent calculations based on the updated population state. The prematurely convergent subpopulation may break its stagnation state due to the loss of its best individual, while the target subpopulation may expand its search direction due to the introduction of new high-quality genes. The candidate matching population composed of all subpopulations is thus updated, providing a key mechanism for solving the problem of design space fragmentation, promoting information exchange and global optimization between different feasible domain fragments.

[0088] Example 2: Figure 2A schematic diagram of a method for intelligent design of low-carbon, high-performance paving material proportions according to the present invention is provided, and a system for intelligent design of low-carbon, high-performance paving material proportions is provided, comprising:

[0089] The constraint acquisition module is used to acquire multi-source constraints of paving materials and construct an initial candidate mix population;

[0090] The constraint coupling module is used to analyze the violation state of the initial candidate mix proportion population under multi-source constraints, and to identify and classify constraint coupling groups that lead to design space fragmentation based on the co-occurrence of violation states.

[0091] The subgroup partitioning module is used to partition the candidate matching population according to the constraint coupling group, and generate and identify the subgroups corresponding to different fragmented feasible domains.

[0092] The potential assessment module is used to calculate the coding entropy of the candidate matching schemes within each subpopulation and the closeness to the average constraint boundary, and to assess the evolutionary potential value of each subpopulation based on the coding entropy and the closeness to the average constraint boundary.

[0093] The independent optimization module is used to allocate optimization computing resources to each subpopulation based on the evolutionary potential value and perform independent optimization calculations, while monitoring and analyzing the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process.

[0094] The population update module is used to perform directional migration operations across subpopulations based on the convergence trend of the evolutionary trajectory, and update the candidate matching population.

[0095] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0096] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0097] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. Computer-readable storage media can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0098] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0099] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0100] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0101] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0102] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0103] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0104] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent design of low-carbon, high-performance paving material proportions, characterized in that, include: S1. Obtain the multi-source constraints of the paving materials and construct an initial candidate mix population; S2. Analyze the violation states of the initial candidate mix ratio population to the multi-source constraints, and identify and classify the constraint coupling groups that lead to the fragmentation of the design space based on the co-occurrence of the violation states. S3. Divide the candidate matching population according to the constraint coupling group, and generate and identify the sub-populations corresponding to different fragmented feasible domains. S4. Calculate the coding entropy of the candidate matching schemes within each subpopulation and the closeness to the average constraint boundary, and evaluate the evolutionary potential of each subpopulation based on the coding entropy and the closeness to the average constraint boundary. S5. Allocate optimization computing resources to each subpopulation based on the evolutionary potential value and perform independent optimization calculations. At the same time, monitor and analyze the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process. S6. Based on the convergence trend of the evolutionary trajectory, perform a directional migration operation for the cross-subpopulation matching scheme, and update the candidate matching population, including: Based on the convergence trend of the evolutionary trajectories of each subpopulation, determine whether there are two subpopulations that tend to be in similar regions in the target space and at least one subpopulation that tends to converge prematurely. When the judgment is yes, the candidate matching scheme with the highest fitness is selected from the subpopulation that tends to converge early as the migrating individuals; Migrating individuals are added to another subpopulation, and the candidate pairing scheme with the lowest fitness in that other subpopulation is replaced to update the candidate pairing population.

2. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 1, characterized in that, S1 includes: Obtain and analyze the multi-source constraints to determine the allowable value range of each component of the paving material; Multiple candidate ratio schemes are randomly generated within the allowed value range; The feasibility of each candidate mix design is determined based on multi-source constraints. Candidate pairing schemes with different feasibility judgment results were selected to form an initial candidate pairing population.

3. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 1, characterized in that, S2 include: For each candidate matching scheme in the initial candidate matching population, record its violation status for each multi-source constraint. Statistical analysis of the co-occurrence frequency of any two multi-source constraints being simultaneously violated by the same batch of candidate allocation schemes; Multi-source constraints with co-occurrence frequencies exceeding a preset frequency threshold are grouped into a constraint coupling group.

4. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 1, characterized in that, S3 include: For each constraint coupling group, analyze the overall satisfaction of the multi-source constraint conditions within the constraint coupling group by each candidate ratio scheme in the candidate ratio population. Candidate matching schemes that are consistent in overall satisfaction are grouped into the same set; Each set generates a subpopulation, and assigns an identifier to the corresponding subpopulation to associate it with its corresponding constraint coupling group.

5. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 1, characterized in that, S4 include: For each subpopulation, the distribution of coding features of candidate pairing schemes within it is statistically analyzed and the coding entropy is calculated; Calculate the distance from each candidate allocation scheme to the boundary of the multi-source constraints, and take the average value to obtain the average constraint boundary proximity. The evolutionary potential value of the subpopulation is calculated by combining the encoding entropy with the average constraint boundary proximity according to the preset rules.

6. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 5, characterized in that, The coding feature distribution of the candidate formulation schemes within the subpopulation is statistically analyzed and the coding entropy is calculated, including: encoding the percentage content of each component in each candidate formulation scheme as a discretized coding bit; statistically analyzing the frequency of occurrence of different discrete values ​​of the coding bit in all coding bits of the subpopulation; and calculating the coding entropy of the subpopulation based on the frequency of occurrence according to the information entropy formula.

7. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 5, characterized in that, The evolutionary potential value of a subpopulation is calculated by combining the encoding entropy and the average constraint boundary proximity according to preset rules. This includes: normalizing the encoding entropy and the average constraint boundary proximity respectively; and weighting the normalized encoding entropy and the normalized average constraint boundary proximity according to preset weights. The result is the evolutionary potential value of the subpopulation.

8. The intelligent design method for the proportioning of low-carbon, high-performance paving materials according to claim 1, characterized in that, S5 include: Each subpopulation is assigned an independent number of iterations based on its evolutionary potential value. Each subpopulation independently updates and filters candidate matching schemes within the allocated number of iterations. During the independent optimization calculation process for each subpopulation, the optimal fitness and coding diversity of each generation of candidate matching schemes are recorded. Based on the recorded changes in optimal fitness and coding diversity over multiple generations, the convergence trend of the evolutionary trajectories of each subpopulation was analyzed.

9. A low-carbon, high-performance paving material proportioning intelligent design system, used to implement the low-carbon, high-performance paving material proportioning intelligent design method according to any one of claims 1-8, characterized in that, include: The constraint acquisition module is used to acquire multi-source constraints of paving materials and construct an initial candidate mix population; The constraint coupling module is used to analyze the violation state of the initial candidate mix proportion population under multi-source constraints, and to identify and classify constraint coupling groups that lead to design space fragmentation based on the co-occurrence of violation states. The subgroup partitioning module is used to partition the candidate matching population according to the constraint coupling group, and generate and identify the subgroups corresponding to different fragmented feasible domains. The potential assessment module is used to calculate the coding entropy of the candidate matching schemes within each subpopulation and the closeness to the average constraint boundary, and to assess the evolutionary potential value of each subpopulation based on the coding entropy and the closeness to the average constraint boundary. The independent optimization module is used to allocate optimization computing resources to each subpopulation based on the evolutionary potential value and perform independent optimization calculations, while monitoring and analyzing the convergence trend of the evolutionary trajectory of each subpopulation during the independent optimization calculation process. The population update module is used to perform directional migration operations across subpopulations based on the convergence trend of the evolutionary trajectory, and update the candidate matching population.

Images